Collection of MCP servers for various APIs and services including Weather, Search, IP lookup, Artifact Registry, Fetch, UFC, and Wikipedia
This server collection exposes 8 tools across multiple fastmcp-based services. Tool naming is consistent and verb-forward (get_, search_, list_), descriptions are present but minimal (most 30-50 chars), and input schemas are properly typed. However, output schemas are NOT documented anywhere in the provided source, parameter descriptions are generic without constraints or examples, and error handling strategies are absent. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are present despite the READ_ONLY risk classification. The Makefile shows HTTP deployment (docker-compose with port exposure), but the actual server implementations (mcp-weather/server.py, etc.) were not provided, so per-tool definitions are partially inferred from the metadata table rather than verified against source code. This inference cap means individual tool scores cannot exceed 50 on schema/description rigor.
Fetch text content from a URL
Get artifact information from Cloud.ru Artifact Registry
Get information about an IP address
Get UFC and MMA fight information
Get weather information for a location
List artifact registries in Cloud.ru
Search the internet using DuckDuckGo
Search Wikipedia for information
Output schemas not documented. LLMs cannot plan downstream calls or extract required fields when tool response structures are unknown.
Descriptions are too brief (30-35 chars). Current descriptions lack WHEN to use each tool and WHAT it returns.
Parameter descriptions are generic or absent. E.g., 'location' in get_weather has description 'Location name or identifier', which does not specify format (city name? coordinates? postal code?) or examples.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 45 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 22 | - | v1 |
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite risk classification showing all tools as READ_ONLY. Modern MCP spec allows agents to understand tool safety profile.
No error handling guidance. Tools do not indicate what errors are possible, which are retryable, or how LLMs should recover (e.g., 'IP lookup failed: try a different format').
search and search_wikipedia are similar but distinct. Tool names do not clearly differentiate when to use DuckDuckGo vs Wikipedia. LLMs will conflate them.
No pagination documented for search and search_wikipedia, which return variable result counts. Without limit/offset/page_size defaults and total_count returns, large result sets will bloat context.
Actual tool implementations (mcp-weather/server.py, etc.) were not provided in source code.